• Panoram'IA: Appointment Friday 15/11 at 10 am
• DSDIR not accessible from the H100 partition
• Shutdown of the old STORE space at the end of November 2024
• IDRIS Training

• Panoram'IA: Appointment Friday 15/11 at 10 am

IDRIS support invites you this Friday morning 15/11 at 10 am to "Panoram'IA": the monthly live video magazine that covers the scientific and technical news of AI. This session's programme includes: AI news, an expert intervention: "QROA: The art and science of jailbreaking LLMs" and our selection of papers with Papers Storm. The live stream and replays are available on our YouTube channel "Un oeil sur l'IDRIS": https://www.youtube.com/@idriscnrs.

• DSDIR not accessible from the H100 partition

We would like to remind you that the DSDIR space, where we store public datasets and models useful to the community (http://www.idris.fr/jean-zay/gpu/jean-zay-gpu-dataset.html), is currently not accessible from the new H100 partition of Jean Zay. Equipment is currently being installed to enable its migration to a new Lustre-based storage system that will be accessible from all partitions. This new space should be operational by the end of 2024. In the meantime, if you need to use a dataset or model available in the DSDIR from an H100 node, we recommend copying it to your SCRATCH.

• Shutdown of the old STORE space at the end of November 2024

Following the migration of the STORE space to a new Lustre-based storage system during the summer, the old STORE space remained accessible in read-only mode using the environment variable "$OLDSTORE" (http://www.idris.fr/jean-zay/modifications-extension-jean-zay-h100.html#cas_particulier_du_store). The definitive shutdown of this space will take place at the end of November 2024.

• IDRIS Training

Remember to register now for the IDRIS training sessions scheduled for the rest of the year and early 2025:

• HPC Debugging, 22 November
• SIMD Vectorisation, 26 November
• Optimised Deep Learning on Jean Zay, 14, 15, 16 and 17 January
• MPI, 18, 19, 20 and 21 March
• OpenMP, 26, 27 and 28 March

For more information on the IDRIS training catalogue and registration procedures: http://www.idris.fr/formations/catalogue.html.

[English version below]

Hello,

Following today's maintenance (Tuesday 1st October), several changes may impact you.

• QoS name changes for the A100 partition

To better manage resource sharing on the machine, specific QoS have been defined for the A100 partition. If you explicitly used the QoS "qos_gpu-t3" or "qos_gpu-dev" in your job submissions targeting this partition, you will need to use "qos_gpu_a100-t3" or "qos_gpu_a100-dev" instead. The QoS "qos_gpu_a100-t3" is used by default and can be omitted.

The CPU and V100 partitions are not affected by this change.

The documentation has been updated accordingly: http://www.idris.fr/jean-zay/gpu/jean-zay-gpu-exec_partition_slurm.html#les_qos_disponibles.

• Use of QoS via JupyterHub

If you wish to specify a QoS when using the Slurm launcher on JupyterHub, you will now need to specify it manually in the "Extra #SBATCH directives" field.

• JupyterHub IP address change

The IP address of our JupyterHub instance has been modified. It is now 130.84.132.56. This change may impact you if your organisation applies IP address filtering for outgoing connections. If you encounter difficulties connecting to JupyterHub, we suggest contacting your IT service to inform them of this change.

As a reminder, the range of IP addresses used for IDRIS machines and services is as follows: 130.84.132.0/23. We recommend authorising the entire range rather than specific IP addresses to avoid being affected by future internal changes to our infrastructure.

• Opening of the H100 partition

Users who have already obtained H100 hours can now use them. You can refer to the example below:

#!/bin/bash
#SBATCH --job-name=mon_travail       # nom du job
#SBATCH -A xyz@h100                  # comptabilite a utiliser, avec xyz le trigramme de votre projet
#SBATCH -C h100                      # pour cibler les noeuds H100
# Ici, reservation de 3x24=72 CPU (pour 3 taches) et de 3 GPU (1 GPU par tache) sur un seul noeud :
#SBATCH --nodes=1                    # nombre de noeud
#SBATCH --ntasks-per-node=3          # nombre de tache MPI par noeud (= ici nombre de GPU par noeud)
#SBATCH --gres=gpu:3                 # nombre de GPU par noeud (max 4 pour les noeuds H100)
# Sachant qu'ici on ne reserve qu'un seul GPU par tache (soit 1/4 des GPUs),
# l'ideal est de reserver 1/4 des CPU du noeud pour chaque tache:
#SBATCH --cpus-per-task=24           # nombre de CPU par tache (1/4 des CPUs ici)
# /!\ Attention, "multithread" fait reference a l'hyperthreading dans la terminologie Slurm
#SBATCH --hint=nomultithread         # hyperthreading desactive

Note that the default modules are not compatible with the H100 partition. To use the software environment specific to this partition, you must load the "arch/h100" module: http://www.idris.fr/jean-zay/cpu/jean-zay-cpu-doc_module.html#modules_compatibles_avec_la_partition_gpu_p6. This must be done in your submission scripts as well as in your terminal if you need to compile codes.

If you do not yet have H100 hours, the project manager can make a request on the eDARI portal if necessary.

Do not hesitate to contact assist@idris.fr if needed.

Best regards, The IDRIS support team

Dear Jean Zay user,

Several changes might affect you after today's maintenance operations (Tuesday October 1st):

• QoS name changes for the A100 partition

In order to more precisely manage the resource sharing of the machine, specific QoS have been defined for the A100 partition. If you used to explicitly specify "qos_gpu-t3" or "qos_gpu-dev" in your Slurm jobs targeting the A100 partition, you now have to use "qos_gpu_a100-t3" or "qos_gpu_a100-dev" instead. Note that the "qos_gpu_a100-t3" QoS is used by default and may be omitted.

The CPU and V100 partitions are not affected by these changes.

The online documentation has been updated: http://www.idris.fr/eng/jean-zay/gpu/jean-zay-gpu-exec_partition_slurm-eng.html#available_qos

• Use of QoS through JupyterHub

If you wish to specify a QoS when using Slurm on JupyterHub, you now have to do it manually in the "Extra #SBATCH directives" field.

• JupyterHub IP address change

The IP address of our JupyterHub instance has been modified. It is now 130.84.132.56. This change might impact you if your institution applies an IP address filtering of outgoing connections. If you run into difficulties when connecting to JupyterHub, we invite you to contact your local administrator to mention this change.

As a reminder, the set of IP addresses used for the IDRIS machines and services is the following: 130.84.132.0/23. We recommend authorising the complete set rather than specific IP addresses so as not to be affected by potential future internal changes of our infrastructure.

• Opening of the H100 partition

Users who were already granted H100 computing hours may now use them. An example submission script is as follows:

#!/bin/bash
#SBATCH --job-name=my_job            # job name
#SBATCH -A xyz@h100                  # account to use, with xyz the 3 letter code of your project
#SBATCH -C h100                      # to target H100 nodes
# Example reservation of 3x24=72 CPU (for 3 tasks) and 3 GPU (1 GPU per task) on one node:
#SBATCH --nodes=1                    # number of nodes
#SBATCH --ntasks-per-node=3          # number of MPI tasks per node (= number of GPU requested per node here)
#SBATCH --gres=gpu:3                 # number of GPU requested per node (max. 4 for H100 nodes)
# Since here only one GPU per task is requested (i.e., 1/4 of the available GPUs)
# the best way to proceed is to book 1/4 of the node's CPU for each task:
#SBATCH --cpus-per-task=24           # number of CPU per task (1/4 of the CPUs here)
# /!\ Caution, "multithread" in Slurm vocabulary refers to hyperthreading.
#SBATCH --hint=nomultithread         # hyperthreading deactived

Note that the default modules are not compatible with the H100 partition. In order to use the software environment dedicated to this partition, you need to load the "arch/h100" module: http://www.idris.fr/eng/jean-zay/cpu/jean-zay-cpu-doc_module-eng.html#modules_compatible_with_gpu_p6_partition. This is needed for your submission scripts but also in your shell when compiling codes.

If you do not have H100 computing hours yet, your project manager may ask for supplementary hours ("au fil de l'eau") on the eDARI portal if necessary.

Do not hesitate to contact assist@idris.fr if needed.

Best regards, The IDRIS support team

[TOOL_CALLS]Arrêt électrique : mardi 8 et mercredi 9 octobre 2024

[English version below]

Bonjour,

Une maintenance de l'infrastructure technique de l'IDRIS nécessitera la mise à l'arrêt du centre mardi 8 et mercredi 9 octobre prochains. La machine Jean Zay sera indisponible du mardi 8 octobre à 6h jusqu'au jeudi 10 octobre à 12h. L'espace STORE sera indisponible dès l'après-midi du lundi 7 octobre.

Le service de support aux utilisateurs sera fermé pendant ces deux jours. Le site web restera accessible et vous pourrez suivre la disponibilité des machines sur la page habituelle : http://www.idris.fr/statut.html.

Nous sommes désolés pour la gêne que cette opération pourrait occasionner.

Cordialement, L'équipe support de l'IDRIS

Hello,

A maintenance operation on the technical infrastructure of IDRIS will require the centre to be shut down completely on Tuesday 8th and Wednesday 9th October. The Jean Zay machine will be unavailable from Tuesday 8th October at 6 am until Thursday 10th October at noon. The STORE disk space will be unavailable starting from Monday 7th October in the afternoon.

User support will be closed for these two days. The website will remain accessible and you will be able to check the machine availability on the usual page: http://www.idris.fr/status.html.

We are sorry for any inconvenience this maintenance operation might cause.

Best regards, The IDRIS User Support Team

• Summary:

• Panoram'IA: Join us on Friday 27/09 at 10 am

• IDRIS Training

• Panoram'IA: Join us on Friday 27/09 at 10 am

IDRIS support invites you this Friday morning 27/09 at 10 am to "Panoram'IA": the monthly live video magazine that covers the scientific and technical news of AI. On the agenda for this session: AI news, TransformerEngine, HyperCloning, and our selection of papers with Papers Storm. The live stream and replays are available on our YouTube channel "Un oeil sur l'IDRIS": https://www.youtube.com/@idriscnrs. The following sessions will take place on 18 October 🎃, 15 November, and 6 December 🎅.

• IDRIS Training

Register now for the IDRIS training sessions scheduled for the rest of the year:

• OpenMP, from 16 to 18 October
• Practical Introduction to Deep Learning, on 4 and 5 November
• Deep Learning Architectures, on 6 and 7 November
• Deep Learning Launch Workshop, on 8 November
• HPC Debugging, on 22 November
• SIMD Vectorisation, on 26 November
• Introduction to the PETSc Library, on 5 and 6 December

For more information on the 2024 IDRIS training catalogue and registration procedures: http://www.idris.fr/formations/catalogue.html.

Opération sur le gestionnaire de travaux Slurm le 1er octobre

[English version below]

Bonjour,

Dans le cadre de l'intégration de la nouvelle partition H100 à la machine Jean Zay, une opération est prévue sur le gestionnaire de travaux Slurm le mardi 1er octobre 2024 de 8h à 13h. La machine sera indisponible pendant cette période.

Contrairement aux opérations de maintenance habituelles, les travaux en attente qui n'auront pas pu s'exécuter avant la maintenance seront perdus. Il vous faudra les soumettre à nouveau au redémarrage de la machine.

De plus, la base de données contenant l'historique des travaux Slurm sera réinitialisée. Vous n'aurez donc plus accès aux informations concernant les travaux antérieurs à la maintenance via la commande "sacct". Néanmoins cela n'impactera pas la comptabilité des heures consommées visible via la commande "idracct".

Nous sommes désolés pour la gêne que cette opération pourrait occasionner.

Cordialement, L'équipe support de l'IDRIS

Hello,

As part of the integration of the new H100 partition to the Jean Zay machine, an operation is planned on the job scheduler Slurm on Tuesday, October 1st 2024 from 8am to 1pm. The machine will be unavailable during that time.

Unlike the usual maintenance operations, the pending jobs that could not run before the maintenance operation will be lost. You will have to resubmit them after the machine is back online.

Moreover, the database which stores the Slurm jobs history will be reset. This means you will not have access anymore to the information about past jobs using the "sacct" command. However, there will be no impact on the accounting of your computing hours visible using the "idracct" command.

We are sorry for any inconvenience this maintenance operation might cause.

Best regards, The IDRIS User Support Team

Summary:​

• Next IDRIS UC Meeting: Thursday 26 September 2024
• IDRIS at the Science Festival on 11 and 12 October 2024
• IDRIS Training

• Next IDRIS UC Meeting: Thursday 26 September 2024

The next User Committee (UC) meeting of IDRIS will take place on 26 September 2024 at the IDRIS premises. Do not hesitate to send your requests to: cu-elus at idris.fr. More information: http://www.idris.fr/cu.html

• IDRIS at the Science Festival on 11 and 12 October 2024

Following the success of previous years, IDRIS is renewing the experience and participating again in the CNRS Unusual Visits organised as part of the Science Festival. Registration is open from 2 to 22 September 2024! For more information: http://www.idris.fr/annonces/idris-fete-de-la-science.html.

• IDRIS Training

Remember to register now for the IDRIS training sessions planned for the rest of the year:

• MPI, from 24 to 27 September
• Jean Zay Workshop, on 3 and 4 October
• OpenMP, from 16 to 18 October
• Optimised Deep Learning on Jean Zay, from 22 to 25 October
• Practical Introduction to Deep Learning, on 4 and 5 November
• Deep Learning Architectures, on 6 and 7 November
• Deep Learning Launch Workshop, on 8 November
• HPC Debugging, on 22 November
• SIMD Vectorisation, on 26 November
• Introduction to OpenACC and OpenMP GPU, from 27 to 29 November
• Introduction to the PETSc Library, on 5 and 6 December

For more information on the 2024 IDRIS training catalogue and registration procedures: http://www.idris.fr/formations/catalogue.html.

Dear Jean Zay user,

As part of the renewal of the storage spaces accompanying the H100 extension of Jean Zay, the current SCRATCH disk space will be permanently shutdown on Tuesday, September 3rd, 2024.

As previously announced, there will be no automatic copy of the current SCRATCH space to the new one.

To anticipate this shutdown, we invite you to migrate as soon as possible the data you wish to keep to the new space already accessible through the $NEWSCRATCH (or $ALL_CCFRNEWSCRATCH for the common project spaces) and to use that disk space in your jobs by modifying your submission scripts accordingly.

The data transfers can be done using one or more jobs on the "archive" partition, or interactively from the login nodes if the volume to be transferred is limited: http://www.idris.fr/eng/jean-zay/modifications-extension-jean-zay-h100-eng.html#scratch_et_all_scratch_copies.

In order to avoid confusions between the different disk spaces and facilitate the monitoring of the migration operation, we suggest that you remove any data that you will not need anymore or that you have already successfully migrated.

Please be aware that the new SCRATCH space is also subject to the 30 day deletion policy of the unused data.

Best regards, The IDRIS support team

Migration of WORK and HOME

[English version below]

Hello,

As previously announced, the arrival of the H100 extension comes with a renewal of the Jean Zay storage spaces with the installation of a new Lustre storage system that will offer increased storage capacity and improved bandwidth.

The migration of the HOME spaces was completed during this morning's maintenance (July 30th, 2024). We invite you to check your scripts to correct any hard-coded paths. Any path of the form "/gpfs7kw/linkhome/..." should become "/linkhome/..." or, if possible, be replaced by the use of the $HOME environment variable.

The migration of the WORK spaces started today. This operation is also handled by the IDRIS teams, so you do not have any specific actions to perform. This operation will be done in batches to avoid a long downtime of the machine. However, it will require suspending the "qos_cpu-t4" and "qos_gpu-t4" QoS, which allow running jobs longer than 20 hours.

For a specific project, the migration process will be as follows:

• 20 hours before the migration begins, jobs using the project's computing hours will no longer be able to start to avoid jobs trying to access the WORK space during the operation (they will then appear with the status "AssocGrpJobsLimit")
• just before the migration starts, the project's WORK space will become completely unavailable, including from the login nodes
• once the migration is completed, the environment variables will be modified to point to the new WORK spaces on the Lustre storage system, and the pending jobs will be able to run again.

Warning: If you have jobs that use the computing hours of one project but access the WORK space of another project, they may fail because we will not be able to block their start appropriately.

A command "idr_migstatus" allows you to monitor the migration of your projects by indicating the current status of each:

• "pending": the migration has not yet been performed, your jobs can still run, and you have access to your WORK
• "planned": the migration will start in the next 24 hours, new jobs can no longer start, but you still have access to your WORK
• "in progress": the migration is in progress, you no longer have access to your WORK
• "migrated": the migration is completed, you have access to your WORK again, and your jobs can run again.

Note: The absolute path of the WORK spaces will change with the migration, but to simplify the transition, links will be set up so that the old absolute paths remain functional, at least initially. Once the migration is completed, we still invite you to modify any paths of the form "/gpfswork/..." or "/gpfsdswork/projects/..." that may appear in your scripts (if possible by replacing them with the use of the environment variable) or in your symbolic links.

We apologise for any inconvenience these operations may cause.

Best regards, The IDRIS support team

Dear Jean Zay user,

As previously announced, the installation on the new H100 extension comes with a renewal of the Jean Zay storage spaces with the installation of a new Lustre storage system offering an increased storage capacity and an improved bandwidth.

The migration of the HOME spaces is completed since today's maintenance operation (July 30th, 2024). We invite you to check your scripts in order to correct any hard-coded paths. Any path starting with "/gpfs7kw/linkhome/..." should become "/linkhome/..." or, if possible, the $HOME environment variable should be used instead.

The migration of the WORK spaces started today. This operation is also handled by the IDRIS teams so you do not have any specific actions to perform. The migration will be done by batch of projects to avoid having a long downtime of the machine. It will however require suspending the "qos_cpu-t4" and "qos_gpu-t4" QoS which allow running jobs of more than 20h.

For a specific project, the migration process will be as follow:

• 20h before the migration begins, the jobs using computing hours allocated to that project will be held in queue (with the "AssocGrpJobsLimit" status) in order to avoid having jobs that use the WORK during the migration operation
• just before the migration starts, the WORK space will become completely unavailable, including from the login nodes
• once the migration is done, the environment variables will be modified to point to the new Lustre WORK space and your jobs will be able to run again.

Warning: If you have jobs that use the computing hours from a project but access the WORK disk spaces of another project, they might fail because we have no way to prevent them from starting when they should not.

The "idr_migstatus" command allows to monitor the migration of your projects by indicating the current status of each of them:

• "pending" : the migration has not started yet, there is no impact on your jobs and you can access your WORK
• "planned" : the migration is going to start in the next 20h, jobs that are not yet running will stay pending but you can still access your WORK
• "in progress" : the migration is in progress, you will not have access to your WORK at this point
• "migrated" : the migration is done, you can access your WORK again and your jobs can run.

Note: The absolute paths of the WORK spaces will be modified by the migration. However to ease the transition, symbolic links will be created in order to keep the old absolute paths working, at least for some times. Once the migration is completed, we do invite you to modify any absolute paths starting with "/gpfswork/..." or "/gpfsdswork/projects/..." that could appear in your scripts (use the environment variables whenever possible) or in your symbolic links.

We are sorry for the inconvenience those operations might cause.

Best regards, The IDRIS support team

SCRATCH Migration Procedure​

[English version below]

Hello,

As previously announced, the arrival of the H100 extension comes with a renewal of the Jean Zay storage spaces with the installation of a new Lustre storage system offering increased storage capacity and improved bandwidth.

To anticipate the shutdown of the current SCRATCH space, we invite you to start using the new space accessible via the environment variable $NEWSCRATCH (or $ALL_CCFRNEWSCRATCH for the common project space) by modifying your submission scripts accordingly. A subsequent communication will specify the date of this shutdown with at least 15 days' notice.

There will be no automatic copying of the current SCRATCH content to the new space. Therefore, if you wish to keep certain data currently stored on the SCRATCH, make sure to copy it to your new space $NEWSCRATCH (for example, via one or more jobs on the "archive" partition or interactively from the login nodes if the volume to be transferred is limited) and delete data you no longer need.

Please note that this new SCRATCH space is also subject to the automatic deletion of data not used for 30 days.

For more information on the ongoing operations on Jean Zay as part of the H100 extension installation: http://www.idris.fr/jean-zay/modifications-extension-jean-zay-h100.html.

Best regards, The IDRIS support team

Dear Jean Zay user,

As previously announced, the installation on the new H100 extension comes with a renewal of the Jean Zay storage spaces with the installation of a new Lustre storage system offering an increased storage capacity and an improved bandwidth.

To anticipate the shut down of the current SCRATCH space, we invite you as of now to start using the new SCRATCH space already accessible through the $NEWSCRATCH (or $ALL_CCFRNEWSCRATCH for the common project spaces) by modifying accordingly your submission scripts. The actual shut down date of the current SCRATCH will be announced later with at least a 15 day notice.

There will be no automatic copy of the current SCRATCH space to the new one. Therefore, if you wish to keep some data currently stored on the SCRATCH, be sure to copy it to your new space $NEWSCRATCH (for instance through one or more jobs on the "archive" partition, or interactively from the login nodes if the volume to be transferred is limited) and to remove data you do not need anymore.

Please be aware that the new SCRATCH space is also subject to the 30 day deletion policy of the unused data.

For more information on the work in progress on Jean Zay for the H100 extension: http://www.idris.fr/eng/jean-zay/modifications-extension-jean-zay-h100-eng.html.

Best regards, The IDRIS support team

Extension of STORE Unavailability​

[English version below]

Hello,

Due to technical difficulties during the migration of the STORE space (see our previous communication for more details on the migration: http://www.idris.fr/flash-info-idris/flash-info-de-l-idris-167.html), the unavailability of this disk space must unfortunately be extended. Access to the STORE should be restored by the end of Thursday, 25 July. We invite you to consult the "Machine availability" page of our website (http://www.idris.fr/status.html) and the message of the day displayed when logging in to Jean Zay for the most up-to-date information.

As a reminder, it is still possible to access the old STORE space in read-only mode using the "$OLDSTORE" environment variable.

We apologise for any inconvenience caused.

Best regards, The IDRIS support team

Dear Jean Zay users,

Due to technical difficulties encountered during the migration of the STORE space (cf. our last email for more details regarding the migration: http://www.idris.fr/flash-info-idris/flash-info-de-l-idris-167.html), this disk space will unfortunately be unavailable for a longer period than expected. Access to the STORE should be possible again by the end of Thursday July 25th. You can check the "Machine availability" webpage (http://www.idris.fr/status.html) and the message of the day displayed when logging in on Jean Zay for the latest updates.

Please note that it remains possible to have read-only access to the old STORE using the "$OLDSTORE" environment variable.

We are sorry for the inconvenience this may cause.

Best regards, The IDRIS support team